Wireless Substitution: State-level Estimates From the National Health Interview Survey, January 2007–June 2010
Number 39 n April 20, 2011 Wireless Substitution: State-level Estimates From the National Health Interview Survey, January 2007–June 2010 by Stephen J. Blumberg, Ph.D., and Julian V. Luke, Division of Health Interview Statistics, National Center for Health Statistics; Nadarajasundaram Ganesh, Ph.D., and Michael E. Davern, Ph.D., NORC at the University of Chicago; and Michel H. Boudreaux, M.S., and Karen Soderberg, M.S., State Health Access Data Assistance Center, University of MinnesotaAbstract Introduction Objectives—This report presents state-level estimates of the percentage of The prevalence and use of wirelessadults and children living in households that did not have a landline telephone telephones (also known as cellularbut did have at least one wireless telephone. National estimates for the 12-month telephones, cell phones, or mobiletime period from July 2009 through June 2010 indicate that 23.9% of adults and phones) has changed substantially over27.5% of children were living in these wireless-only households. Estimates are the past decade. Today, an ever-also presented for selected U.S. counties and groups of counties, for other increasing number of adults have chosenhousehold telephone service use categories (e.g., those that had only landlines to use wireless telephones rather thanand those that had landlines yet received all or almost all calls on wireless landline telephones to make and receivetelephones), and for 12-month time periods since January–December 2007. calls. As of the first half of 2010, more Methods—Small-area statistical modeling techniques were used to estimate than one in four American householdsthe prevalence of adults and children living in households with various household (26.6%) had only wireless telephones—telephone service types for 93 disjoint geographic areas that make up the entire an eightfold increase over just 6United States. This modeling was based on January 2007–June 2010 data from years (1). The prevalence of suchthe National Health Interview Survey, 2006–2009 data from the American ‘‘wireless-only’’ households nowCommunity Survey, and auxiliary information on the number of listed telephone markedly exceeds the prevalence oflines per capita in 2007–2010. households with only landline Results—The prevalence of wireless-only adults and children varied telephones (12.9%), and this differencesubstantially across states. State-level estimates for July 2009–June 2010 ranged is expected to grow.from 12.8% (Rhode Island and New Jersey) to 35.2% (Arkansas) of adults and The increasing prevalence offrom 12.6% (Connecticut and New Jersey) to 46.2% (Arkansas) of children. For wireless-only households hasadults, the magnitude of the increase from 2007 to 2010 was lowest in New implications for telephone surveys.Jersey (7.2 percentage points) and highest in Arkansas (14.5 percentage points). Many health surveys, political polls, andKeywords: cell phones • telephone surveys • noncoverage • small domain other research studies are conductedestimation using random-digit-dial (RDD) telephone surveys. Until recently, these surveys did not include wireless telephone numbers in their samples. Now, despite operational challenges, most major survey research organiza tions include wireless telephone U.S. DEPARTMENT OF HEALTH AND HUMAN SERVICES Centers for Disease Control and Prevention National Center for Health Statistics
Page 2 National Health Statistics Reports n Number 39 n April 20, 2011numbers when conducting RDD are needed. Direct state-level estimates Methodstelephone surveys. If they did not, the of this prevalence have not beenexclusion of households with only available from NHIS data because the Small-area statistical modelingwireless telephones (along with the sample size of NHIS is insufficient for techniques were used to combine NHIS2.0% of households that have no direct, reliable annual estimates for most data collected from within specifictelephone service) could bias states. However, in March 2009 NCHS geographies (states and some counties)results (2–4). released the results of statistically with auxiliary data that are repre Best practices for conducting modeled estimates of the prevalence of sentative of those geographies tosurveys by calling wireless telephones wireless-only adults at the state level, produce model-based estimates.are not yet known, but substantial using data from the 2007 NHIS and the Specifically, we used a combination ofresearch has been conducted to address direct survey estimates from the 2008 Current Population Survey’sthe known operational challenges (5). 2007–2010 NHIS, direct survey Annual Social and EconomicStatistical challenges also exist when estimates from the 2006–2009 ACS, and Supplement (6). Those estimates werecombining samples of wireless-only auxiliary information on the number of the first state-level estimates of the sizehouseholds with samples of landline listed telephone lines per capita inhouseholds from RDD surveys. To of this population available from the 2007–2010. The small-area model wasensure that each sample is appropriately federal government. used to derive estimates of therepresented in the final data set and In this report, we update those proportion of people who lived inappropriately weighted in the final original 2007 estimates. We present households that were wireless-only,analyses, reliable estimates of the results of modeled estimates of the wireless-mostly, dual-use, landlineprevalence of wireless-only households prevalence of wireless-only adults and mostly, and landline-only for theare needed (5). Moreover, if the persons wireless-only children at the state level, following seven 6-month periods:interviewed on their wireless telephones using data from the 2007–2010 NHIS January–June 2007, July–Decemberare not screened to exclude persons who and the 2006–2009 American 2007, January–June 2008, July–also have landlines, reliable estimates of Community Survey (ACS), along with December 2008, January–June 2009,the prevalence of landline and wireless auxiliary information on the number of July–December 2009, and January–telephone service use may be required listed telephone lines per capita. By June 2010.to address the probability that an incorporating data from multiple Estimates were derived for adultsindividual could be in both samples (5). sources, the modeled estimates presented and children for 93 nonoverlapping The National Health Interview here take advantage of the unique areas that make up the entire UnitedSurvey (NHIS) is the most widely cited strengths of each data set. States. Twenty-six of these areas weresource for data on the ownership and This report also expands on the states, and one was the District ofuse of wireless telephones. Every 6 original 2007 estimates in three Columbia; others areas consisted ofmonths, the Centers for Disease Control important ways. First, it includes selected counties, groups of counties, orand Prevention’s (CDC) National Center estimates for 42 additional substate the balance of the state populationfor Health Statistics (NCHS) releases a excluding the selected counties. No geographic areas in the United States.report with the most up-to-date areas crossed state lines, and every Second, it includes estimates not onlyestimates available from the federal location in the United States was part of for July 2009–June 2010, but also forgovernment concerning the size and one (and only one) of the 93 areas. 12-month time periods from Januarycharacteristics of the wireless-only Areas considered for inclusion in thispopulation (1). That report, published as 2007 through June 2010. Third, it report included urban areas that receivepart of the NHIS Early Release includes estimates not only for persons federal Section 317 immunization grantsProgram, presents national and regional living in wireless-only households, but and other substate areas that are strataestimates. also for additional household telephone for CDC’s National Immunization Many RDD telephone surveys are service use categories. Estimates are Survey (7). Areas were selected for thisdesigned to collect data and produce presented for adults and children living report on the basis of available surveyresults at the state or local level, in wireless-mostly households (defined sample sizes and the stability of theincluding several surveys conducted by as households that have landlines yet modeled estimates.CDC (e.g., the Behavioral Risk Factor receive all or almost all calls on For each telephone category, theSurveillance System, the National wireless telephones), dual-use 6-month estimates for all 93 small areasImmunization Survey, and the State and households (which receive significant were modeled jointly. That is, allLocal Area Integrated Telephone amounts of calls on both landlines and 6-month periods were modeled togetherSurvey). For such surveys to effectively wireless telephones), landline-mostly in a single model rather than separatelycombine samples of wireless-only households (which have wireless as seven models (one for each 6-monthhouseholds with samples of landline telephones yet receive all or almost all period). Separate small-area modelshouseholds, state-level estimates of the calls on landlines), and landline-only were fitted for each telephone serviceprevalence of wireless-only households households. use category (e.g., wireless-only,
National Health Statistics Reports n Number 39 n April 20, 2011 Page 3dual-use) and by age group (adults or these three distinct sets of estimates 12-month state-level estimates werechildren). The model-based estimates for were combined, the weights associated obtained by appropriately weighting theeach telephone service use category, with each set reflected the relative 12-month small-area estimates bysmall area, and 6-month period were precision of each estimate. population size.derived using a standard small-area Although model-based estimates Further detail regarding thismodeling and estimation approach were produced for every small area and estimation methodology is available inknown as ‘‘empirical best linear 6-month period, consecutive 6-month the Technical Notes section.unbiased prediction’’ (8–10). The period estimates were combined tomodel-based estimates were a weighted produce 12-month estimates. The Estimates for Adultscombination of three distinct sets of small-area estimates for 12-month and Children Livingestimates: (a) the direct estimate from periods were obtained by averaging two in Wireless-onlyNHIS for the small area during the consecutive 6-month estimates. This6-month period of interest, (b) a helped to reduce the variability of the Householdssynthetic estimate derived from a estimates. Then, the 12-month small- Results from the small-arearegression model involving ACS and area estimates for each phone category modeling strategy showed greatother auxiliary data for the small area were adjusted so that they agreed with variation in the prevalence of adultsduring the 6-month period of interest, the national direct estimates from NHIS living in wireless-only householdsand (c) ‘‘adjusted direct estimates’’ from for the corresponding phone category across states (Figures 1 and 2).NHIS for the small area during all and year. The 12-month estimates were Estimates for July 2009–June 20106-month periods other than the 6-month further adjusted so that they agreed with ranged from a high of 35.2% inperiod of interest. By using estimates the 2008 or 2009 ACS estimate for the Arkansas to a low of 12.8% in Rhodefrom all seven 6-month periods, the population with a telephone (either Island and New Jersey (Table 1). Othermodel-based estimate allows for landline or wireless) for each small area. states in which the prevalence of‘‘borrowing strength’’ across time. When For states with multiple small areas, wireless-only adults was relatively high Less than 20% 20% to less than 25% 25% to less than 30% Greater than or equal to 30% DATA SOURCES: CDC/NCHS, National Health Interview Survey, January 2007–2010; U.S. Census Bureau, American Community Survey, 2006–2009; and infoUSA.com consumer database, 2007–2010. Estimates were calculated by NORC at the University of Chicago.Figure 1. State-level comparisons of the percentage of adults living in wireless-only households, using modeled estimates: United States,July 2009–June 2010
Page 4 National Health Statistics Reports n Number 39 n April 20, 2011 40 35 30 25 Percent 20 15 10 5 0 WA WY WV NM MO MN MD AR OR OK GA OH CO SC SD ND WI NC MS NH NE TN TX NV ME CA NY MA DE MT UT CT KY KS AZ VA AK PA VT AL LA FL NJ ID IN MI HI RI IA IL State DATA SOURCES: CDC/NCHS, National Health Interview Survey, January 2007–2010; U.S. Census Bureau, American Community Survey, 2006–2009; and infoUSA.com consumer database, 2007–2010. Estimates were calculated by NORC at the University of Chicago.Figure 2. Modeled state-level estimates of the percentage of adults living in wireless-only households: United States, July 2009–June 2010(exceeding 31%) were Mississippi low prevalence of wireless-only children (14.1) and North Dakota (13.0). Other(35.1%), Texas (32.5%), North Dakota included New Hampshire (15.0%), states with a smaller-than-average(32.3%), Idaho (31.7%), and Kentucky Massachusetts (15.1%), Rhode Island increase included New York (7.3),(31.5%). Several other states in the (15.8%), and New York (16.6%). Pennsylvania (7.5), Rhode Island (7.5),Northeast region joined Rhode Island Table 1 also provides the modeled and Utah (7.6). Table 2 can be used toand New Jersey with prevalence rates estimates of the prevalence of wireless- produce similar estimates of changebelow 17%, including Connecticut only adults for each 12-month time over time for children living in wireless-(13.6%), New Hampshire (16.0%), period from January 2007 through June only households.Pennsylvania (16.5%), Delaware 2010. Nationally, the prevalence of(16.5%), and Massachusetts (16.8%). wireless-only adults increased from Estimates for AdultsPrevalence rates were also relatively low 13.6% to 23.9%, an absolute increase of Living in Householdsin South Dakota (15.6%). 10.3 percentage points. As expected, the With Wireless Similarly, results showed great values increased in every state fromvariation in the prevalence of wireless- 2007 to 2010, and the increase in Telephonesonly children across states, ranging from prevalence was statistically significant in Table 3 presents modeled estimatesa high of 46.2% in Arkansas to a low of every state. The absolute increase from for July 2009–June 2010 for the12.6% in Connecticut and New Jersey 2007 to 2010 ranged from a high of prevalence of adults living in(Table 2). Other states with a high 14.5 percentage points in Arkansas to a households with various telephoneprevalence of wireless-only children low of 7.2 percentage points in New service types, including but not limitedincluded Mississippi (41.9%), North Jersey. Other states with a larger-than- to wireless-only status. Estimates areDakota (39.7%), New Mexico (38.9%), average increase in the prevalence of presented for adults living in wireless-and Idaho (37.3%). Other states with a wireless-only adults included Mississippi mostly households, landline-mostly
National Health Statistics Reports n Number 39 n April 20, 2011 Page 5households, dual-use households, and wireless-only persons (11). The results Pennsylvania), Davidson Countylandline-only households. These results in this report clearly show that, for (Nashville, Tennessee), Dallas Countycan be used to obtain the prevalence of many states, national and regional (Dallas, Texas), and King Countyadults living in households with any estimates are not sufficiently accurate (Seattle, Washington), where thewireless telephones (regardless of for these purposes. prevalence of wireless-only adultswhether the wireless telephones are the Results from the small-area significantly exceeded the correspondingonly telephones). Estimates ranged from statistical models show great state-level state-level prevalence.a high of 91.8% in Iowa to a low of variation in the prevalence of wireless- Prevalence estimates are included47.9% in South Dakota. Other states only adults, even within regions. The not only for July 2009–June 2010, butexceeding 90% included Utah (90.9%), range of prevalence exceeded also for 12-month time periods fromColorado (90.7%), Kansas (90.7%), 10 percentage points in the Northeast January 2007 through June 2010. TheMinnesota (90.3%), and Delaware region, 13 percentage points in the West statistical model based on 3½ years of(90.3%). Other states below 70% region, 16 percentage points in the data—and therefore larger sample sizesincluded Montana (60.6%), Wyoming Midwest region, and 18 percentage in each geographic area—is more stable(63.3%), and Nevada (66.2%). points in the South region. In fact, in than a model based on only a single Table 3 can also be used to look at the Midwest region, the state with the year of data. Estimates from the morethe prevalence of adults living in lowest prevalence (South Dakota, stable model are presumed to be morehouseholds that receive all or almost all 15.6%) borders the state with the reliable. Thus, we presume that thecalls on wireless telephones, regardless highest prevalence (North Dakota, estimates for 2007 presented in thisof whether the households have landline 32.3%). Wider ranges within regions report are more reliable than thetelephones. Both wireless-only and were observed for estimates of the estimates for 2007 presented in ourwireless-mostly adults are in this group. prevalence of wireless-only children. previous report (6). Modeled estimatesEstimates of the prevalence of adults Survey researchers and for January 2007–June 2009 forliving in households where wireless telecommunications companies household telephone service usetelephones are the primary means of interested in local areas may question categories other than wireless-only havereceiving calls ranged from 52.8% in whether state-level prevalence estimates not been included in this report but areTexas to 24.9% in South Dakota. Other are sufficiently specific. This report available upon request.states exceeding 47% included Arkansas includes estimates for 42 counties or The estimates developed for this(50.9%), Mississippi (49.8%), Arizona groups of counties, selected from a list report are based on data from 2007(48.1%), and Nebraska (47.3%). Other of immunization-policy-relevant areas through 2010. The number of Americanstates below 30% included Connecticut on the basis of available survey sample homes with only wireless telephones(28.2%), New Hampshire (29.4%), and sizes and the stability of the modeled continues to grow (1), and it is veryRhode Island (29.6%). estimates. Most of these substate areas likely that the current prevalence rates Table 4 presents modeled estimates are major metropolitan cities, and of wireless-only adults and children arefor July 2009–June 2010 for the national estimates suggest that adults greater than the estimates presentedprevalence of children living in living in metropolitan areas are more here. Researchers may find that the rateshouseholds with various telephone likely to live in wireless-only of growth presented in Tables 1 and 2service types. The table can be used to households than are adults living in for states and substate areas are usefulcalculate estimates for children similar nonmetropolitan areas. The mean of the for predicting current or futureto those for adults described above. 42 substate-area estimates of the prevalence rates. prevalence of wireless-only adults Finally, the state and substate (26.7%) was greater than the mean of estimates presented here may differDiscussion the ‘‘rest of state’’ estimates for those 24 from estimates produced by other Because of the limited availability states (23.5%). However, for the sources. For example, Arbitron, Inc.,of reliable and updated state-level majority of the substate areas, the released Fall 2009 estimates of theprevalence estimates for the wireless- prevalence of wireless-only adults did prevalence of wireless-only householdsonly population, survey researchers not differ significantly from the area’s in local radio markets (12). Theirinterested in combining state-level corresponding state-level prevalence estimates are based largely on surveysamples of wireless-only households estimate. Exceptions included Orange responses received from mailedwith samples of landline households County (Orlando, Florida), Cook County screening questionnaires, which may behave relied on national or regional (Chicago, Illinois), Madison/St. Clair subject to various nonresponse biases.estimates of the relative sizes of these counties (Metro East St. Louis, Illinois), The estimates presented here are lesstwo populations (5). Similarly, Marion County (Indianapolis, Indiana), likely to be biased by surveytelecommunications companies seeking Suffolk County (Boston, Massachusetts), nonresponse (due to the high NHISgreater understanding of conditions in Wayne County (Detroit, Michigan), response rates), but are more likely tostate and local markets have relied on Essex County (Newark, New Jersey), be biased by the focus here onregional estimates of the prevalence of Allegheny County (Pittsburgh, demographic characteristics in the
Page 6 National Health Statistics Reports n Number 39 n April 20, 2011statistical model. Arbitron’s estimates 7. CDC. National Immunization Survey:reveal higher wireless-only prevalence A user’s guide for the 2008 public-estimates in areas with college campuses use data file. 2009. Available from:or military bases; the statistical models ftp://ftp.cdc.gov/pub/here did not include any community Health_Statistics/NCHS/characteristics. The NHIS sample also Dataset_Documentation/NIS/ NISPUF08_DUG.pdf.does not include active-duty military 8. Jiang J, Lahiri P. Mixed modelpersonnel. Survey nonresponse, sample prediction and small area estimationcharacteristics, and model selection (with discussion). Test 15(1):1–96.should all be considered when 2006.evaluating or comparing small-area 9. Rao JNK. Small area estimation.estimates, including those presented Hoboken, NJ: John Wiley & Sons.here. 2003. 10. Rao JNK, Yu M. Small areaReferences estimation by combining time-series and cross-section data. Can J Stat1. Blumberg SJ, Luke JV. Wireless 22(4):511–28. 1994. substitution: Early release of estimates based on data from the 11. Petitions of Qwest Corporation for National Health Interview Survey, forbearance pursuant to 47 U.S.C. January–June 2010. National Center §160(c) in the Denver, Minneapolis- for Health Statistics. December 2010. St. Paul, Phoenix, and Seattle Available from: http://www.cdc.gov/ metropolitan statistical areas. WC nchs/nhis.htm. Docket No. 07–97, Memorandum2. Blumberg SJ, Luke JV. Coverage Opinion and Order, FCC 08–174. bias in traditional telephone surveys Federal Communications of low-income and young adults. Commission. July 25, 2008. Public Opin Q 71(5):734–49. 2007. Available from: http://hraunfoss.3. Blumberg SJ, Luke JV, Cynamon fcc.gov/edocs_public/attachmatch/ ML. Reevaluating the need for FCC-08-174A1.pdf. concern regarding noncoverage bias 12. Arbitron, Inc. Cell-phone-only in landline surveys. Am J Public household penetration in Arbitron Health 99(10):1806–10. 2009. radio metro areas. 2009. Available4. Blumberg SJ, Luke JV, Cynamon from: http://www.arbitron.com/ ML, Frankel MR. Recent trends in downloads/cell_phone_penetration household telephone coverage in the _map.pdf. United States. In: Lepkowski JM, 13. Jiang J, Lahiri P, Wan S-M, Wu C-H. Tucker C, Brick JM, et al., eds. Jackknifing in the Fay-Herriot model Advances in telephone survey with an example. In: Statistical methodology: 56–86. Hoboken, NJ: John Wiley & Sons. 2007. policy working paper 33:5. AAPOR Cell Phone Task Force. New Proceedings of the Seminar on considerations for survey researchers Funding Opportunity in Survey when planning and conducting RDD Research; 98–104. Washington, DC: telephone surveys in the U.S. with Federal Committee on Statistical respondents reached via cell phone Methodology, Office of Management numbers [online]. American and Budget. 2001. Association for Public Opinion 14. Magee L. R2 measures based on Research. 2010. Available from: Wald and likelihood ratio joint http://aapor.org/cell_phone_ significance tests. Am Stat task_force.htm. 44(3):250–3. 1990.6. Blumberg SJ, Luke JV, Davidson G, et al. Wireless substitution: State- level estimates from the National Health Interview Survey, January– December 2007. National health statistics reports; no 14. Hyattsville, MD: National Center for Health Statistics. 2009. Available from: http://www.cdc.gov/nchs/data/nhsr/ nhsr014.pdf.